Research map / Label-efficient and robust learning / Semi-supervised learning
Semi-Supervised Risk Control via Prediction-Powered Inference
Bat-Sheva Einbinder, Liran Ringel, Yaniv Romano
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use CLT upper confidence bound instead of finite-sample bounds | SS-RCPS · ImageNet APS calibration (130 labeled) | coverage violation rate: 18% instead of 10% | Hurt |
| use WSR finite-sample bound instead of CLT bound | SS-RCPS · ImageNet APS calibration (small labeled set) | coverage conservativeness | Hurt |
| use specialized binary-loss semi-supervised RCPS instead of vanilla RCPS lab | RCPS · ImageNet APS calibration (small labeled set) | coverage / set size | Improved |
| use WSR semi-supervised RCPS instead of vanilla RCPS and BIN as labeled size grows | RCPS · ImageNet APS calibration (large labeled set) | coverage / set size | Improved |
| impute unlabeled samples naively without correction (RCPS imp) | RCPS · ImageNet APS calibration | coverage validity | Hurt |
| increase imputation accuracy of unlabeled labels | SS-RCPS BIN · ImageNet APS calibration (130 labeled) | coverage variance and set size | Improved |
| tune power parameter lambda with CLT UCB (ablation) | SS-RCPS CLT · ImageNet APS calibration | set size | Improved |
| tune lambda via data splitting with WSR UCB at high accuracy (ablation) | SS-RCPS WSR · ImageNet APS calibration (2000 labeled) | statistical efficiency | Hurt |
| add semi-supervised calibration with BIN at 300 labeled samples | RCPS lab · IMDb early time series classification (300 labeled) | halt time and variance | Improved |
| add semi-supervised calibration with BIN at 4000 labeled samples | RCPS lab · IMDb early time series classification (4000 labeled) | halt time | No clear effect |
| use WSR semi-supervised RCPS at 300 labeled samples | RCPS lab · IMDb early time series classification (300 labeled) | halt time | Hurt |
Closest papers by meaning
- Active, anytime-valid risk controlling prediction sets (NeurIPS 2024 · similarity 0.91)
- Performative Risk Control: Calibrating Models for Reliable Deployment under Performativity (NeurIPS 2025 · similarity 0.86)
- Provably Reliable Conformal Prediction Sets in the Presence of Data Poisoning (ICLR 2025 · similarity 0.85)
- Semi-Supervised Regression with Heteroscedastic Pseudo-Labels (NeurIPS 2025 · similarity 0.85)
- Keep It on a Leash: Controllable Pseudo-label Generation Towards Realistic Long-Tailed Semi-Supervised Learning (NeurIPS 2025 · similarity 0.84)
- Prediction-Powered Active Testing (NeurIPS 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
